Article 4RKAQ Skydio’s New Drone Is Smaller, Even Smarter, and (Almost) Affordable

Skydio’s New Drone Is Smaller, Even Smarter, and (Almost) Affordable

by
Evan Ackerman
from IEEE Spectrum on (#4RKAQ)

When Skydio announced the R1 in early 2018, it was one of the most incredible drones we'd ever seen. It's been a year and a half, and in the fast-paced world of drones, the Skydio R1 is somehow still, by a huge margin, the most intelligent and capable drone in existence, offering a level of autonomy that would be impressive even if it was a one-off research project, which it wasn't, because you could buy one for US $2,500.

The R1, though, was really not intended to be a consumer drone in the sense that it wasn't a direct competitor to the likes of DJI, which has overwhelmingly dominated the consumer drone space since the early days of consumer drones. Rather, the R1 was meant to demonstrate exactly what Skydio was capable of, offering the chosen few who could justify paying for one a magical experience that couldn't be found anywhere else.

Today, Skydio is announcing their second drone: the Skydio 2. The Skydio 2 takes everything that made the R1 so amazing, and squeezes it into something smaller, smarter, and at $999, alarmingly close to affordable.

Video: Skydio

The Skydio 2 is a drone with built-in subject tracking and obstacle avoidance with omnidirectional coverage, including above and below. Using an array of cameras, the drone can follow people or vehicles at speed while nimbly dodging anything that might get in the way, all while recording smooth, cinematic video-and you don't have to do anything besides telling the drone who you want it to follow. Other drones claim similar functionality, but generally only under ideal and very limited conditions. Skydio's drones may not be able to handle everything, but they can handle almost everything, which is saying a lot when you consider the inherently three dimensional and unstructured and cluttered up world that they have to operate in. The Skydio 2 represents autonomy that you can trust.

How Skydio 2's obstacle avoidance works

While the launch video above does a good job of highlighting the Skydio 2's subject tracking, you only get a few hints of just how impressive the obstacle avoidance capability is. Wide open spaces are dramatic and all, but the brief flight through that warehouse shows the kind of autonomy that Skydio's drones excel at. This is basically state-of-the-art for drones right now, and it's in a consumer product that costs under $1,000. If that original video of the Skydio R1 chasing a mountain biker through a forest blew you away (which it totally should have), we're told that the Skydio 2's autonomy is better. Way, way better.

Video: Skydio This video shows the feed from one of the three top navigation cameras [left circle] and the deep learning predicted depth map [right circle] as a Skydio 2 drone flies through three challenging scenarios. The first has a lot of clutter and repeated texture (trees and branches); the second has thin lines (poles with overhanging wires); and the last has large textureless surfaces indoors. The final part of the video shows the 3D point cloud of the drone flying around Skydio's office.

The most obvious change between the Skydio R1 and Skydio's new drone is that the structure that wrapped all the way around the R1 and contained 12 VGA cameras has been completely eliminated. The Skydio 2 uses just six cameras, each of which is a 4K fisheye with a 200-degree field of view, which was tricky for Skydio to implement successfully, as Skydio CEO Adam Bry explains:

That was a super challenging engineering problem, because we went from these conventional global shutter cameras to rolling shutter fisheye cameras, which are very very challenging from an algorithmic standpoint. But the big advantage is that at a lower price point, we can get much better sensor performance. The simple metric on this is we went from about 3 megapixels of visual sensing on R1, to about 45 megapixels on Skydio 2.

The placement of these cameras resulted in the drone's curious over/under propeller design, which ensures that the cameras on the top and bottom of the drone get a complete hemispherical view unobstructed by propellers. In fact, the Skydio 2 can see nearly every point in space with three cameras at once, resulting in "better than stereo" depth perception.

MzM4NTc1NA.gif Gif: Skydio Skydio's autonomy system is powered by an NVIDIA Tegra X2 processor running nine custom deep networks. The drone uses a 3D world model with an update rate of more than 1 million points per second, and is able to track up to 10 simultaneous objects.

In order to deal with the amount of data required to successfully dodge obstacles at its top speed of 58 km/h (36 mph), the Skydio 2 is packing a very tightly integrated NVIDIA Tegra X2 inside of its magnesium frame. The frame actually functions as a heat sink, conducting heat away from the processors out underneath the propellers, which act as a very effective cooling system. The autonomy system is still sucking down a lot of power, though-Bry says that the drone's compute budget is about 10 percent of its total power consumption, meaning that there's a couple minutes of flight time that you're not getting because the drone is constantly thinking so hard.

Controller and beacon: extended range MzM4NTU5Ng.jpeg Image: Skydio In addition to an app for iOS and Android, Skydio now offers a dedicated physical controller with a 3.5-kilometer range if you'd like to fly it by hand. To be clear, "by hand" doesn't deactivate the obstacle avoidance, and the drone will fly itself around obstacles and keep you from running into things quite effectively.

Since the Skydio 2 is primarily intended to fly, rather then be flown, it doesn't come standard with a physical controller. The good news is that Skydio does now offer a dedicated physical controller for the drone for an extra $150, if you'd like to fly it by hand. To be clear, "by hand" doesn't deactivate the obstacle avoidance, and the drone will fly itself around obstacles and keep you from running into things quite effectively.

MzM4NTU5OA.jpeg Image: Skydio You can use the beacon to give simple commands to the drone, but its primary purpose is to allow the drone to track you when you're out of sight (the beacon has a 1.5 km range and uses GPS and GLONASS).

If the controller looks familiar (and it should), that's because it's a Parrot Skycontroller3 (which ships with the Parrot Anafi drone) with Skydio's logo on the front. There's nothing sneaky going on here-Skydio is not particularly interested in making controllers, so they're buying Skycontroller3s directly from Parrot and loading in their own software. You can read more about the Skycontroller3 as part of my Anafi review, but basically, it's a solid, simple controller that's a little chunky but has an excellent antenna.

The other accessory that you can buy for $150 is a wireless beacon. You can use the beacon to give simple commands to the drone, but its primary purpose is to allow the drone to track you when you're out of sight, which is important if you want to take advantage of the Skydio 2's obstacle avoidance as much as possible. While the communication range between a Skydio 2 and a smartphone is about 200 meters, the beacon extends that to 1.5 kilometers, and the controller has a range of 3.5 km.

Who should buy a Skydio 2?

We recognize that at $1,000 plus accessories, calling this drone "affordable" could be a stretch for most people. But if you're in the market for a new drone of any kind, the Skydio 2 is worth a serious look. To put the Skydio 2's cost in perspective with what could conceivably be called its main-stream competition: Parrot's Anafi drone is currently about $650, features limited tracking capability but no obstacle avoidance. You can find a DJI Mavic Air for about $750, which does include some fairly primitive obstacle avoidance from the front. A DJI Mavic 2, which (on paper) purports to offer a somewhat similar omnidirectional obstacle detection and avoidance capability as Skydio, starts at around $1,400, but we can't emphasize enough that Skydio's autonomy is in a totally and completely different class to anything you'd experience with DJI.

The Skydio 2 also holds up well when compared across other metrics. Independent testing (commissioned by Skydio, so take it with a grain of salt) demonstrates that the Skydio 2's 12-megapixel main camera (which also shoots 4K video) can hold its own against the Mavic 2's camera. With the optional controller, you'll have no problem flying the Skydio 2 farther than you can see, and 23 minutes of flight time is respectable. While the 775-gram Skydio 2 weighs somewhere in between the Mavic Air and Mavic 2 Pro, it doesn't fold down, although it is quite flat with the 4280 mAh battery taken out.

MzM4NTU4Mg.jpegMzM4NTYxNg.jpeg Photos: Skydio The Skydio 2 doesn't fold down but is quite flat with the battery taken out. The drone almost certainly won't fit in your pocket, but you'd have no trouble fitting its hard case into a small bag, so for most practical purposes it's similarly portable to a folding drone, weighing in at 775 grams.

Since some of Skydio 2's navigation cameras are on its arms, adding a folding mechanism (and making those arms less rigid) would have resulted in significantly degraded performance, we're told. The drone almost certainly won't fit in your pocket, but you'd have no trouble fitting its hard case into a small bag, so for most practical purposes it's similarly portable to a folding drone.

Essentially, the Skydio 2 is offering a magical amount of autonomy and no significant downsides for a premium of a few hundred dollars over what you'd pay for a much (much) dumber drone. You're not just paying for something that's marginally better, you're paying for access to entirely new capabilities that manage to deliver on what drones have been promising for half a decade. If your primary goal is to learn how to pilot a drone, perhaps the Skydio 2 isn't for you. But if your primary goal is to use a drone, that's what the Skydio 2 is all about: taking amazing pictures and videos of yourself and your friends and family without having to worry about the process of not crashing into things and keeping your subject in frame. And let's face it, unless you're already an expert pilot, the Skydio 2 is going to do a much better job anyway.

But while the Skydio 2 is a "consumer" drone, it's certainly not limited to being used by consumers, and Skydio is already thinking about the advantages that a fully autonomous drone could offer to industry. You don't have to think very hard to see how big of a deal this is-if you replace "autonomous filming" with "autonomous data collection" there are all kinds of immediate applications. Skydio has an SDK that it's opening up to commercial partners, starting with DroneDeploy, a mapping company. And Skydio promises that over the next several months, the Skydio 2 will get to work in everything from infrastructure inspection to assisting first responders.

Video: Skydio The placement of Skydio 2's navigation cameras resulted in the drone's curious over/under propeller design, which ensures that the cameras on the top and bottom of the drone get a complete hemispherical view unobstructed by propellers. Designing Skydio 2

For more details about Skydio and their new drone, we spoke with Skydio CEO Adam Bry.

IEEE Spectrum: How did developing the R1 help you get to the Skydio 2?

Adam Bry: When we started designing R1, we were basically five people in a house. The team was super small. There was a ton of stuff we didn't know, and nobody had ever built a product like that before. Because of that, it was a real venture into the unknown. We got a lot of stuff right, but we also learned a huge amount through that development process that there was just no way that we'd have been able to capture in a first product. The other side of it is that with the development of R1, we basically weren't able to optimize for cost at all-we optimized for trying to make the thing work, because it was so fundamentally challenging.

Why is autonomy so important for drones?

Our basic belief is that drones are like other computing devices-phones and laptops and tablets-where the right device with the right core technology built into it becomes a platform for many different use cases. We think autonomy is the thing that enables that, because autonomy is what turns it into a software-defined experience. Autonomy is what makes it possible to write applications for different use cases. Without autonomy, the application is in the skill of the pilot, and I don't think that's long-term scalable for drones.

How has Skydio's autonomy gotten better?

The starting point is having much more data coming in and a faster computer, which opens up some new possibilities, but even with all the hardware stuff, the biggest progress is still on the software side. One of the big things that we've done is come up with an entirely new depth perception algorithm based on deep learning rather than more conventional [photogrammetric] stereo techniques, which generally work in simple scenes but fail when there's a lot of repeated structure or textureless surfaces.

What people do is combine contextual scene understanding with photometric information-when we look at a scene, we know what everything is and generally how the world is structured around us. And the combination of that contextual information with the photometric information means that people and animals essentially have perfect depth perception based just on vision. We've come up with a deep learning algorithm that mimics this, such that the drone not only uses photometric information but also contextual information. This drastically improves performance in every scenario that we care about, and overcomes a lot of the really tricky cases. For example, flying around in a big textureless room, we see near perfect performance. And with power lines, if it can see any part of the power line, it does the same thing that a person does and assumes that the line extends. It's a really, really powerful thing.

"One of the big things that we've done is come up with an entirely new depth perception algorithm based on deep learning rather than more conventional [photogrammetric] stereo techniques, which generally work in simple scenes but fail when there's a lot of repeated structure or textureless surfaces" -Adam Bry, Skydio

Is the drone aware of its own confidence level in its autonomy, and does it communicate that to the user?

There are some scenarios where it doesn't know enough to be cautious, but everywhere we can, we bake caution into the software. A couple of tangible examples: It checks the lightning conditions and will warn you and come to a hover if there's just not enough light to fly. It also detects smudges or other damage to the lenses or cameras, and behaves conservatively-if it has a smudge in a particular direction, it'll try not to fly in that direction. The way that we perceive the environment, we have a notion of uncertainty baked in, and so if the drone is seeing things that it's never seen before and can't make sense of what's going on, it knows that and degrades gracefully while letting the user know what's happening.

This is why we put the safe operating guidelines out there. We don't yet claim that these things are uncrashable, because we don't want people to try. In normal use, everything should be fine, but the more drones we ship, the more likely it is that someone will find something we've never seen before. Those situations usually turn out to be useful learning experiences for us that we use to improve and update the software.

MzM4NTU5NQ.jpeg Photo: Skydio The Skydio 2 drones are hand made in California. According to the company's warranty, if you're operating your Skydio 2 within its Safe Flight guidelines and it crashes, Skydio will repair or replace it for free.

What's the best way to get more of these advanced drone capabilities into consumer products?

One thing I would say, and this may be controversial in some circles, is I think that autonomous drones, and full fledged autonomous robots in general, have become systems engineering challenges. A really interesting thing that's happened over the last five years is that the most advanced systems have transitioned from being built in academia to being built by companies, because the technology is now mature enough, and there are real markets for it, meaning that you can make significant R&D investments and build strong teams across a bunch of different disciplines.

I've seen this firsthand: I was a grad student at MIT from 2009 to 2012, and when we were building our systems there, it was basically two of us trying to do everything across hardware and software and embedded firmware and it was all kind just barely held together with chewing gum and duct tape. At Skydio, we have a world class team across hardware and software and they're all working super closely together with aligned goals, and that's something you can't replicate in a research lab.

In your press release, you referred to the drone industry as "not healthy." Can you elaborate on that?

I think the drone industry is still largely a story of unmet potential. Drone generated enormous hype and excitement five or six years ago, and for the most part, I think we haven't realized most of the concepts and use cases to the level that was promised. That gap is one of the things that motivated us to start the company, and our belief is that autonomy is a huge part of what makes these things possible.

That's one angle. The other angle is the blunt fact that DJI has been so dominant, and controls so much of the market, that I think it's resulted in an unhealthy dynamic. It stops suppliers from wanting to invest in supporting technologies, it frustrates customers who want an alternative, and we think this creates an opportunity for us.

The last piece of it, which is something that is getting more and more attention, is that I think people have realized that drones are really important for national security. These consumer devices that a lot of people started off thinking were just toys are actually in a lot of ways the most cutting-edge aerospace systems out there, and have raced beyond a lot of the capabilities of more conventional military drones. And whatever you think about US-China relations, it's probably not great for long-term US national security if the US isn't capable of manufacturing drones and leading the way in drone technology.

A limited number of Skydio 2 drones (which are hand made in California) are available for pre-order as of right now, and a $100 deposit will get your name on the list for delivery starting in November.

[ Skydio ]

VuRd4Y5PJ5k
External Content
Source RSS or Atom Feed
Feed Location http://feeds.feedburner.com/IeeeSpectrum
Feed Title IEEE Spectrum
Feed Link https://spectrum.ieee.org/
Reply 0 comments